注意力机制下的VMD-IDBiGRU负荷预测模型

VMD-IDBiGRU Load Prediction Model under Attention Mechanism

针对传统神经网络在负荷预测中精度欠佳、预测速度较慢的问题,提出一种基于注意力机制、变分模态分解和改进的深度双向门控循环单元短期负荷预测模型。该模型首先通过变分模态分解算法将负荷数据分解,以降低原始负荷数据的复杂度。然后:针对传统分解加预测组合模型存在参数多、模型复杂的问题,基于权值共享的思想建立改进的深度双向门控循环单元神经网络;引入注意力机制来突出关键因素的影响,通过注意力权重深度挖掘负荷数据存在的规律。最后,以中国某地区的负荷数据作为实例,通过与传统预测模型进行对比得出,本文所提模型在精度和速度方面均有一定的提升。

To address the problems in load prediction by traditional neural networks such as poor accuracy and slow pre⁃diction speed,a short-term load prediction model based on attention mechanism,variational modal decomposition(VMD)and improved deep bi-directional gated recurrent unit(GRU)is proposed.First,the load data is decomposed by the VMD algorithm to reduce the complexity in the original load data.Then,to address the problems of many parame⁃ters and model complexity in the traditional combined decomposition and prediction model,an improved deep bi-directional GRU neural network is established based on the idea of weight sharing.In addition,the attention mechanism is introduced to highlight the influence of key factors,and the attention weights are used to deeply mine the laws existing in the load data.The proposed model is compared with the traditional prediction model with the load data of one region in China as an example,indicating that the accuracy and speed of the novel model are improved.